emlearn-micropython is a Machine Learning and Digital Signal Processing for MicroPython. Provides convenient and efficient MicroPython modules, and enables MicroPython application developers to run efficient Machine Learning models on microcontroller, without having to touch any C code.

What's Changed

Functionally identically to 0.10.0. Just a few metadata updates for the Zenodo archival.

New Contributors

• @nkrusch made their first contribution in #64

Full Changelog: 0.10.0...v0.10.1

emlearn-micropython is a Machine Learning and Digital Signal Processing for MicroPython. Provides convenient and efficient MicroPython modules, and enables MicroPython application developers to run efficient Machine Learning models on microcontroller, without having to touch any C code.

Deprecation notice

Using build files from master is deprecated and will be removed in a release or two. Switch to use latest or a version/tag intead.

What's Changed

• Builds for most recent version are now provided for latest (latest release) instead of master.
• Update to Micropython 1.27 by @jonnor in #62
• Add initial Webassembly build by @jonnor in #61

Internal changes

• har_trees: Generalize preprocessing support by @jonnor in #57
• Bump Mac OS in CI to 26 Tahoe by @jonnor in #58

Full Changelog: 0.9.0...0.10.0

What's Changed

New features

• Add Linear Regression, including on-device learning by @jonnor in #43
• Native modules now available for armv6m (RP2040 / Cortex M0 / Cortex M3) and rv32imc (RISC-V / ESP32-C5 / ESP32-C3 / ESP32-C2)
• neighbors: Support building as user module by @jonnor in #47

Internal changes

• tests: Improve running by @jonnor in #55
• Builds with standard MicroPython from master/1.26.1, no patches needed

Full Changelog: 0.8.0...0.9.0

What's Changed

Convolutional Neural Network

• CNN: Build both fp32 and int8 modules by @jonnor in #29

New Documentation

Detailed documentation is now available at https://emlearn-micropython.readthedocs.io/en/latest/

Support for building as external C module (experimental)

• Extmod: Merge .py module code for trees, setup Unix build and tests in CI by @jonnor in #37
• extmod: Add FFT, arrayutils and CNN by @jonnor in #38
• extmod: Start support for cmake RP2040/ESP32 by @jonnor in #39
• extmod: Add more modules to cmake ports by @jonnor in #40
• Initial testing building as C module by @jonnor in #36

Internal changes

• Use new linking support by @jonnor in #31

Full Changelog: 0.7.0...0.8.0

Prebuilt binary modules available at https://github.com/emlearn/emlearn-micropython/tree/gh-pages/builds/0.7.0

What's Changed

• Update to MicroPython 1.24 by @jonnor in #21
• cnn: Fix hardcoded outputs size leading to corrupted predictions by @jonnor in #23
• CNN: Fix wrong output values due to missing dequant by @jonnor in #25
• trees: Switch to output prediction probabilities by @jonnor in #26
• Example for Human Activity Recognition improved. Including data recording functionality and training custom models https://github.com/emlearn/emlearn-micropython/tree/master/examples/har_trees

Full Changelog: 0.6.0...0.7.0

Compatible with MicroPython 1.23.
Pre-built modules are available at https://github.com/emlearn/emlearn-micropython/tree/gh-pages/builds/0.6.0

What's Changed

• Add arrayutils with linear_map function by @jonnor in #16
• Make all modules have emlearn_xxx names by @jonnor in #17
• Large improvements to multiple examples, including har_trees and soundlevel_iir

Full Changelog: 0.5.0...0.6.0

No functional changes over 0.1.0

Testing Zenodo integration

Supports RandomForest/DecisionTree